Welcome to PracticeUpdate! We hope you are enjoying temporary access to this content.
Please register today for a free account and gain full access
to all of our expert-selected content.
Already Have An Account? Log in Now
Clinical Grade Detection of Microsatellite Instability in Colorectal Tumors by Deep Learning
TAKE-HOME MESSAGE
- In routine hematoxylin and eosin (H&E) histological images, microsatellite instability (MSI) and deficient mismatch repair (dMMR) tumors are characterized by distinct morphological patterns such as tumor-infiltrating lymphocytes, mucinous differentiation, heterogeneous morphology, and poor differentiation. A deep learning model reliably differentiated samples with a mean area under the ROC curve of 0.95; with color normalization, this further improved to 0.96.
- Analyzing data from more than 8000 patients with colorectal cancer in an international consortium, this study demonstrates that deep learning can reliably detect MSI and dMMR tumors based on routine H&E histology alone.
– Omar Khan, MD
abstract
This abstract is available on the publisher's site.
Access this abstract now Full Text Available for ClinicalKey SubscribersBACKGROUND AND AIMS
Microsatellite instability (MSI) and mismatch-repair deficiency (dMMR) in colorectal tumors are used to select treatment for patients. Deep learning can detect MSI and dMMR in tumor samples on routine histology slides faster and cheaper than molecular assays. But clinical application of this technology requires high performance and multisite validation, which have not yet been performed.
METHODS
We collected hematoxylin and eosin-stained slides, and findings from molecular analyses for MSI and dMMR, from 8836 colorectal tumors (of all stages) included in the MSIDETECT consortium study, from Germany, the Netherlands, the United Kingdom, and the United States. Specimens with dMMR were identified by immunohistochemistry analyses of tissue microarrays for loss of MLH1, MSH2, MSH6, and/or PMS2. Specimens with MSI were identified by genetic analyses. We trained a deep-learning detector to identify samples with MSI from these slides; performance was assessed by cross-validation (n=6406 specimens) and validated in an external cohort (n=771 specimens). Prespecified endpoints were area under the receiver operating characteristic (AUROC) curve and area under the precision-recall curve (AUPRC).
RESULTS
The deep-learning detector identified specimens with dMMR or MSI with a mean AUROC curve of 0.92 (lower bound 0.91, upper bound 0.93) and an AUPRC of 0.63 (range, 0.59-0.65), or 67% specificity and 95% sensitivity, in the cross-validation development cohort. In the validation cohort, the classifier identified samples with dMMR with an AUROC curve of 0.95 (range, 0.92-0.96) without image-preprocessing and an AUROC curve of 0.96 (range, 0.93-0.98) after color normalization.
CONCLUSIONS
We developed a deep-learning system that detects colorectal cancer specimens with dMMR or MSI using hematoxylin and eosin-stained slides; it detected tissues with dMMR with an AUROC of 0.96 in a large, international validation cohort. This system might be used for high-throughput, low-cost evaluation of colorectal tissue specimens.
Additional Info
Clinical-Grade Detection of Microsatellite Instability in Colorectal Tumors by Deep Learning
Gastroenterology 2020 Jun 17;[EPub Ahead of Print], A Echle, HI Grabsch, P Quirke, PA van den Brandt, NP West, GGA Hutchins, LR Heij, X Tan, SD Richman, J Krause, E Alwers, J Jenniskens, K Offermans, R Gray, H Brenner, J Chang-Claude, C Trautwein, AT Pearson, P Boor, T Luedde, NT Gaisa, M Hoffmeister, JN KatherFrom MEDLINE®/PubMed®, a database of the U.S. National Library of Medicine.
Artificial intelligence (AI) is the ability of computers to perform a task associated with intelligent beings, including the capability to learn. Machine learning (ML) is a branch of AI concentrating on computer algorithms that can learn from data and perform specific tasks and analyses, often without specific human programming. Deep learning (DL) is an advanced subclass of machine learning that relies on specific algorithms called artificial neural networks.
Over the last few years, AI has increasingly been placed at the center of the presumptive next wave of technology that should revolutionize how many aspects of our lives are run. The medical field, and more specifically, gastroenterology, has not been spared from this trend. In fact, several studies have addressed the use of AI in aspects such as assisting colonoscopy for adenoma detection, differentiating diminutive adenomas from hyperplastic polyps, predicting GI bleed, and staging gastric cancer.
In this study by Echle et al, the authors tried to determine the potential usefulness of DL in order to assess microsatellite instability (MSI) status in different sets of colorectal tumors. This has become an issue of most importance, as it has been widely recommended to assess MSI as a way to prescreen patients for genetic testing to rule out Lynch syndrome. Furthermore, MSI status has become relevant for chemotherapy treatment choices due to the unique response of MSI tumors to immune checkpoint inhibitors, not only in colorectal adenocarcinomas but also in all other solid tumors.
Although the study showed significant diagnostic accuracy, it also underlined the complexity of such analyses when applied to different cohorts. There were also important differences in the performance associated such aspects like anatomical location of the tumor (possibly related to prior treatment received for rectal cancers), immune infiltration, or differentiation status of the tumor. Nevertheless, this is likely just the beginning of how AI can impact multiple medical aspects that can have major clinical and economic implications.